Tenement: Process hypervisor for single-server multi-tenancy
ToolsComments
Does the "under a second" wake time include the time for the process to actually handle requests, or just the launch? Cold start latency usually varies significantly depending on the runtime being wrapped.
If we pair this with something like llmlite or other tiny Zig binaries... would that make the scale-to-zero basically instantaneous? I wonder if this is the ideal setup for those ultra-lightweight gateways...
Regardless of binary size, the OP is right about the Kubernetes tax. A control plane on a single VPS often consumes more memory than the actual workloads.
The binary size is one thing, but routing is the real hurdle. Previous attempts at this failed because they didn't integrate the routing layer, making the subdomain routing mentioned in the post the most critical detail.